An integrated optimization and deep learning pipeline for predicting live birth success in IVF using feature...

Area of Science:

• Reproductive Medicine
• Artificial Intelligence in Healthcare
• Biomedical Data Science

Background:

• Predicting in vitro fertilization (IVF) success is challenging due to complex clinical, demographic, and procedural factors.
• Accurate prediction of live birth outcomes is crucial for optimizing assisted reproductive technology treatments.

Purpose of the Study:

• To develop a highly accurate artificial intelligence (AI) pipeline for predicting live birth outcomes in IVF.
• To enhance the interpretability of AI models used in fertility treatment prediction.

Main Methods:

• Evaluated various feature selection methods (PCA, PSO) and machine learning classifiers (RF, Decision Tree, Transformer, Tab_transformer).
• Analyzed confounding factors (age, previous cycles) and preprocessing techniques.
• Employed Shapley Additive Explanations (SHAP) for model interpretability.

Main Results:

• The combination of Particle Swarm Optimization (PSO) for feature selection and a Tab_transformer deep learning model achieved 97% accuracy and 98.4% AUC.
• SHAP analysis identified key predictors of infertility and improved model interpretability.

Conclusions:

• Developed a robust AI pipeline for predicting IVF live birth outcomes with high accuracy and interpretability.
• The AI pipeline shows potential for enhancing personalized fertility treatments and improving patient care.